The Claude Desktop Field Guide

The installable Claude app for macOS and Windows, alongside web and mobile.

Use when Use it when you want a persistent window separate from your browser, or when a feature needs access to something on your machine that the web app can't reach.

A single back-and-forth thread with Claude — the basic unit of using the product.

Use when Use a fresh chat for a new topic; keep going in the same chat while the earlier context still matters.

The message you send Claude — the clearer and more specific, the better the reply.

Use when Every turn. Invest more structure in the prompt when the task is ambiguous, high-stakes, or has a required output format.

The specific Claude version answering you — a trade-off of capability, speed, and cost.

Use when Reach for the most capable model on genuinely hard reasoning, analysis, or coding; use a faster model for quick, well-scoped tasks.

Caps on how much you can use Claude in a window, varying by plan and model.

Use when Worth understanding before a big session so you can pace heavy work, and worth checking if Claude tells you a limit was reached.

Everything Claude can see for the current reply: the thread, files, and connected knowledge.

Use when Think about context whenever answers drift: too little and Claude lacks the facts; too much unrelated material and it loses focus.

Attach a document, image, or spreadsheet to a message so Claude can work with it.

Use when Use it whenever the task is about a specific document or image you have on hand.

Find and pull context from earlier conversations across your history.

Use when Use it when you know you discussed something before but cannot remember which chat, or want to continue earlier work.

Lets Claude carry useful context forward across chats so you repeat yourself less.

Use when Useful when you want continuity across many chats; less appropriate when you want a strictly clean slate (use Incognito chat then).

A one-off conversation that is kept out of your history and Memory.

Use when Use it for a throwaway question, a sensitive one-off, or when you deliberately want a clean slate with no carry-over.

A mode that lets Claude reason for longer before answering, for harder problems.

Use when Turn it on for problems where getting the reasoning right matters more than speed; leave it off for quick lookups or simple rewrites.

The links Claude shows for facts it pulled from the web or documents.

Use when Look at them any time accuracy matters — especially for anything you'll act on, publish, or repeat to someone else.

Lets Claude look things up on the live web and cite what it found.

Use when Use it for recent events, current prices or facts, or anything that may have changed since the model was trained.

Lets Claude open a specific URL you give it and read that page.

Use when Use it when you already have the URL and want Claude to read that specific page rather than go searching.

Claude runs many searches across sources to produce a cited, structured report.

Use when Use it for a question that needs breadth and synthesis — a market scan, a literature-style overview, comparing many options — rather than a single fact.

A container that gives a set of chats shared knowledge and standing instructions.

Use when Use one for any body of work you return to repeatedly — a client, a codebase, a course, an ongoing report — where the same background applies.

Reference files added to a Project that every chat in it can draw on.

Use when Use it for reference material that many chats need: the same brand guide, the same product spec, the same dataset.

Standing directions applied to every chat in a Project.

Use when Use them to encode "how I always want Claude to behave here": the voice of a brand, a coding style, a reviewer persona, a required format.

Claude builds real, downloadable files — documents, spreadsheets, slides — for you.

Use when Use it when you need a deliverable in an editable, shareable file format rather than text in the chat.

Claude runs code in a sandbox to compute, analyse data, and build files.

Use when Use it for data analysis, calculations, transformations, or anything where you want a verifiable computed result rather than an estimate.

A standalone piece of content Claude renders in a side panel you can edit and reuse.

Use when Artifacts appear for content worth working on as an object: something long, reusable, or that you will revise, rather than a quick inline answer.

An artifact you publish to a link so others can view or interact with it.

Use when Use it to hand a small tool, page, or document to a colleague or client without exporting and emailing files.

A tailored visual Claude designs in chat or Cowork — diagrams, layouts, graphics.

Use when Use them when you need a visual explanation or a designed graphic and want it shaped to your specific content.

An interactive artifact in Cowork that runs and updates as you use it.

Use when Use it when you want a small working tool or interactive view — a calculator, a dashboard-like widget, an interactive explainer — inside Cowork.

Claude revises an existing artifact or document directly instead of restating it.

Use when Use it whenever you are refining a draft: "tighten the intro", "change the tone of section two", "fix this function".

Earlier states of an artifact you can review and restore.

Use when Use it whenever you make a series of edits and want a safety net, or want to compare an earlier draft with the current one.

A workspace where Claude works on multi-step tasks with tools, files, and approvals.

Use when Use it for work that is more than a single answer: a task with several steps, tool use, and a deliverable at the end.

One unit of work in Cowork — a task Claude carries out with its own state and outputs.

Use when Frame each distinct piece of delegated work as its own task, so its scope, permissions, and results stay contained.

Lets Claude read and work with files in a folder on your own machine.

Use when Use it when the task is about files that live on your machine and you would rather grant a folder than upload files one by one.

A Cowork task that runs remotely rather than on your local machine.

Use when Use it for longer or scheduled work, or when you do not want or need the task to touch your local files.

A Cowork task set to run automatically on a schedule.

Use when Use it for anything you would otherwise do on a regular cadence: a morning digest, a weekly status draft, a routine check.

A helper agent Claude spins up to handle part of a larger task in parallel.

Use when Relevant for larger Cowork tasks that naturally break into parallel pieces; usually orchestrated by Claude rather than something you invoke directly.

How much Claude asks before acting in Cowork — from confirm-everything to run-freely.

Use when Use more manual approval for unfamiliar, sensitive, or irreversible work; loosen it only once you trust the task and its scope.

A Chrome extension that lets Claude see and act in your browser to help with web tasks.

Use when Use it for tasks rooted in the browser: working through a web app, gathering information across pages, or filling in a repetitive web workflow.

Lets Claude control your computer — screen, clicks, typing — to carry out tasks in Cowork.

Use when Use it for tasks that span apps or GUIs that have no other integration — where the only way to do the job is to actually drive the interface.

A packaged set of instructions and resources that teaches Claude a specific job well.

Use when Use a Skill for a recurring, well-defined job you want done the same way every time — a formatting standard, a domain workflow, a specialised procedure.

A bundle of capabilities you add to Cowork to extend what Claude can do there.

Use when Use plugins to give Cowork extra, task-specific powers beyond the built-in tools, without wiring things up yourself.

A link between Claude and an external app or data source so it can use that data.

Use when Use a connector when the material Claude needs lives in another system you would rather connect than export and upload.

A connector that can also take actions in a service, not just read from it.

Use when Use it when you want Claude not just to see data in a service but to do something there — with appropriate care, since actions have consequences.

One place to browse and add Skills, Connectors, and Plugins.

Use when Go here whenever you want to add a capability, connect a service, or discover what extensions exist.

Standing information and preferences about you that Claude applies across chats.

Use when Set them for preferences that hold everywhere: your role, your tone preferences, formatting you always want.

A reusable preset for how Claude writes — tone, length, and format.

Use when Use a style to match the moment: concise for quick answers, formal for a report, casual for a note.

The controls over which tools and data Claude may use, and how far it can act.

Use when Set and review them any time you connect data, enable a capability, or run agentic work — especially for anything sensitive.

Connects Claude to your Microsoft 365 so it can work with your Office content.

Use when Set it up if your documents and email live in Microsoft 365 and you want Claude to work with them.

Claude working inside Microsoft Word to draft and edit documents.

Use when Use it when you are writing or editing in Word and want Claude's help without leaving the document.

Claude working inside Microsoft Excel to analyse and build spreadsheets.

Use when Use it for spreadsheet work: understanding a complex formula, transforming data, or drafting a model.

Claude working inside Microsoft PowerPoint to build and refine slides.

Use when Use it when building or polishing a deck and you want help without leaving PowerPoint.

Claude working inside Microsoft Outlook to help with email.

Use when Use it for email work: catching up on a long thread, drafting a considered reply, or triaging messages.

Lets Claude work with your files stored in SharePoint and OneDrive.

Use when Use it when the documents Claude needs live in SharePoint or OneDrive rather than on your device.

Claude's edits shown as reviewable suggestions you accept or reject.

Use when Use them any time edits matter enough to review — contracts, published copy, anything where a silent change could slip through.

The open standard many connectors use to link Claude to external tools and data.

Use when Relevant when you want to integrate a custom or third-party system, or understand how connectors work beneath the surface.

A connector you set up yourself to link Claude to a system not in the directory.

Use when Use it when the integration you need is not in the directory and you can point Claude at an appropriate server or endpoint.

A connector whose server runs locally on your own machine.

Use when Use it for integrations that should stay on your machine, or for local tools and data you do not want to expose to a hosted service.

A connector whose server is hosted remotely rather than on your machine.

Use when Use it for services that live in the cloud, or integrations a team shares centrally.

A packaged extension that installs a local integration into the desktop app.

Use when Use it when a capability is distributed as a desktop extension and you want an installable rather than hand-configured integration.

What a tool or connector is allowed to reach over the network — a key safety control.

Use when Relevant when configuring connectors, custom integrations, or agentic environments, especially in organisations with security requirements.

Fetching relevant material and feeding it to the model so answers are grounded in it.

Use when Conceptually relevant whenever Claude answers from Project knowledge, connected data, or uploaded files — retrieval is what pulls the right material in.

The maximum amount of text a model can consider at once, measured in tokens.

Use when Matters on long chats, big documents, and large tool outputs, where you risk pushing important material out of the window.

Artifacts kept beyond a single message so you can return to and reuse them.

Use when Relevant whenever you build something in an artifact you intend to come back to or share.

An interactive artifact that itself calls Claude to power its behaviour.

Use when Relevant when you want to build a small tool whose functionality depends on the model doing work at run time, not just fixed logic.

Points where custom automation can run in response to events, mainly in Claude Code.

Use when A developer-level control for enforcing checks, formatting, or policy automatically around what Claude does in a coding workflow.

Where plugins are distributed and discovered for you to add.

Use when Go here to find plugins that extend Cowork for a kind of work, or to see what the ecosystem offers.

A Skill provided and managed centrally by your organisation.

Use when Relevant in a team or company setting where standardised, vetted Skills are rolled out to everyone.

Admin-set controls over what members can use, connect, and do in Claude.

Use when They apply whenever you use Claude through an organisation; they explain why a feature you have heard of may be unavailable or need approval.

A feature released for early use that may change, break, or be limited.

Use when Worth trying to get capability early, with the understanding that behaviour and availability may shift.

An early, experimental release to gather feedback — expect more limits than a beta.

Use when Try one to explore where Claude is heading, with the expectation that it is rough and evolving.

Anthropic's agentic coding tool that works in your terminal and editor.

Use when Use it for real software work: navigating a codebase, making multi-file changes, running tests, and automating development tasks.